gmaas

Generative Models as a Service allows users to get the models they need for their specific use case trustlessly.

Key Features:

1. Decentralized Marketplace: Enables AI developers to offer their models for use and monetize their work.
2. Zero-Knowledge Proofs (ZKML): Ensures model privacy and verifies that the outputs are generated by the chosen model.
3. EIP7007: A draft NFT standard for AI generated content , drafted by members of PSE in the EF who research on ZKML.
4. Unified Platform: Provides a wide variety of generative models in a single platform for easy accessibility.
5. On-Chain Transactions: Ensures trust and transparency by handling all transactions and proofs on-chain.

Why? For model architects: Direct monetization, model privacy, and access to users For users: Trustable outputs, easy access to a variety of models, pay as you go

How?

1. Users get one prompt to try all of the models
2. Users deposit money
3. Users choose their prompt and the model they want
4. Model architects generate the image, NFT, and the proof of model output.
5. Model architects send the proof to the model to withdraw the money.

The system collects user's signed prompt from the frontend end, and pass to the backend which hosts ML models to generates picture.

The each model correspond to one EIP 7007 NFT on chain, which is and registered on our Marketplace contract when model vender deploys the model. The NFT is minted to user withe a Zero Knowledge proof to guarantee the integrity of model inference.

We use EZKL to generate the proof in our backend, all proofs needs to be verified before mint. EZKL is a ZKML framework based on Halo2 proof system and it generates SNARK verifier as a smart contract. There exists the following correspondence: Model <=> EIP7007 NFT <==> Verifier Contract.